Characterization of tumor endothelial cells (TEC) in gastric cancer and development of a TEC-based risk signature using single-cell RNA-seq and bulk RNA-seq data

Abstract

Background: Tumor endothelial cells (TECs) are essential participants in tumorigenesis. This study is focused on elucidating the TEC traits in gastric cancer (GC) and constructing a prognostic risk model to predict the clinical outcome of GC patients.

Methods: Single-cell RNA sequencing (scRNA-seq) data were obtained from the GEO database. Using specific markers, the Seurat R package aided in processing scRNA-seq data and identifying TEC clusters. Based on TEC cluster-associated genes identified by Pearson correlation analysis, TEC-related prognostic genes were screened by lasso-Cox regression analysis, thereby constructing a risk signature. A nomogram was created by combining the risk signature with clinicopathological features.

Results: Based on the scRNA-seq data, 5 TEC clusters were discovered in GC, with 3 of them showing prognostic associations in GC. A total of 163 genes were pinpointed among 3302 DEGs as significantly linked to TEC clusters, leading to the formulation of a risk signature comprising 8 genes. Furthermore, there was a notable correlation between the risk signature and the immune cell infiltration. Multivariate analysis findings indicated that the risk signature served as an independent prognostic factor for GC. Moreover, its efficacy in forecasting immune response was validated.

Conclusion: TEC-based risk model is highly effective in predicting the survival outcomes of GC patients and can forecast the immune response. Targeting TECs may significantly inhibit tumor progression and enhance the efficacy of immunotherapy.

Keywords: gastric cancer; immunotherapy; risk signature; tumor endothelial cells.

Figure 1

The flow chart of this study.

Figure 2

The identification of TEC clusters based on scRNA seq data of GC patients. (A) UMAP plot of the distribution of 18 samples. (B) UMAP plot of the distribution of five TEC clusters after clustering. (C) Dot plot of the top 5 marker gene expression of TEC clusters. (D) The proportion of the five TEC clusters in tumor samples and normal samples. (E) KEGG enrichment analysis of 5 TEC clusters. (F) UMAP distribution map of malignant and non-malignant cells predicted by Copykat package.

Figure 3

The characteristics of tumor-related pathways in TEC clusters. (A) Heatmap of 10 tumor-related pathway scores enriched in TEC cells. (B) Comparison of TEC clusters in malignant and non-malignant cells. (C–G) Comparison of GSVA score of each pathway between malignant and non-malignant cells in TEC_0 (C), TEC_1 (D), TEC_2 (E), TEC_3 cluster (F), and TEC_4 cluster (G). ^*P < 0.05, ^**P < 0.01, ^***P < 0.001, ^****P < 0.0001, Abbreviation: ns: not significant.

Figure 4

The associations between the five TEC cluster and prognosis of GC patients. (A–E) Comparison of five TEC scores in cancer and normal tissues, ^**P < 0.01, ^***P < 0.001, ^****P < 0.0001, Abbreviation: ns: not significant. (F–J) K-M curves of the high and low TEC score groups in the TEC_0 cluster (F), TEC_1 cluster (G), TEC_2 cluster (H), TEC_3 cluster (I), and TEC_4 (J).

Figure 5

Identification of the hub genes to construct a risk signature. (A) Volcano plot of differentially expressed genes of cancer and normal tissues in TCGA cohort. (B) GO analysis. (C) KEGG analysis. (D) Volcano plot of prognosis-related genes identified from univariate Cox regression analysis. (E) The trajectory of each independent variable with lambda. (F) Plots of the produced coefficient distributions for the logarithmic (lambda) series for parameter selection (lambda). (G) The multivariate Cox coefficients for each gene in the risk signature. (H) K-M curves of risk model constructed by 8 genes in TCGA cohort. (I) ROC curves of risk model constructed by 8 genes in TCGA cohort. (J) K-M curves of risk model constructed by 8 genes in GSE62254 cohort. (K) ROC curves of risk model constructed by 8 genes in GSE62254 cohort. (L) K-M curves of risk model constructed by 8 genes in GSE15459 cohort. (M) ROC curves of risk model constructed by 8 genes in GSE15459 cohort.

Figure 6

The development of a nomogram for predicting the prognosis of GC. (A, B) Univariate and multivariate Cox analysis of risk score and clinicopathological characteristics. (C) Nomogram model integrating the risk score and T stage, N stage was constructed. (D) Calibration curves for 1, 3 years of nomogram. (E) Decision curve for nomogram. (F) Comparison of predictive capacity of clinicopathological features and the nomogram using time-ROC analysis. ^***P < 0.001.

Figure 7

The characteristics of mutations of the genes included in the risk signature. (A) Waterfall diagram of SNV mutations of 8 key genes. (B) Colinearity and mutual exclusion analysis of 8 key genes and the 10 most mutated genes in tumors. (C) CNV mutations (gain, loss, none) of 8 key genes. (D) Correlation heatmap of 8 key genes with Aneuploidy Score, Homologous Recombination Defects, Fraction Altered, Number of Segments, and Nonsilent Mutation Rate.

Figure 8

Identification of the pathways in which the risk genes are implicated. (A) Heatmap showing the correlation between genes and pathways. (B) Heatmap displaying the enrichment scores for key pathways. ^*P < 0.05, ^**P < 0.01, ^***P < 0.001.

Figure 9

The relationship between the risk genes and immune landscape. (A, B) The correlation matrix of the risk genes and StromalScore, ImmuneScore, and ESTIMATEScore. (C) Comparison of high and low expression of 8 key genes and ImmuneScore. (D) Correlation between 8 key genes and immune cell score predicted by CIBERSORT analysis. (E) Correlation between 8 key genes and 10 immune cell types predicted by MCPcounter analysis. ^*P < 0.05, ^**P < 0.01, ^***P < 0.001, and ^****P < 0.0001.

Figure 10

The response of risk score to immune checkpoint inhibitors in IMvigor210 cohort and GSE78220 cohort. (A) Prognostic differences among risk score groups in the IMvigor210 cohort. (B) Differences in risk scores among immunotherapy responses in the IMvigor210 cohort. (C) Distribution of immunotherapy responses among risk score groups in the IMvigor210 cohort. (D) Prognostic differences between risk score groups in early-stage patients in the IMvigor210 cohort. (E) Prognostic differences between risk score groups in advanced patients in the IMvigor210 cohort. (F) Prognostic differences in risk score groups in the GSE78220 cohort. (G) Differences in risk scores among immunotherapy responses in the IMvigor210 cohort. (H) Distribution of immunotherapy responses among risk score groups in the GSE78220 cohort. ^*P < 0.05, ^***P < 0.001.